Signed network models for portfolio optimization

2025-10-06

AI Summary

Traditional portfolio optimization inadequately captures negative correlation structures, leading to suboptimal systemic risk mitigation. Method: We propose a signed-weighted network–based asset selection framework: (i) construct a signed correlation network from a denoised correlation matrix—where positive/negative edges represent positive/negative return correlations—and (ii) apply a discrete optimization algorithm to identify a high-quality asset subset exhibiting strong negative interdependencies; portfolio weights are then assigned via an integrated Markowitz mean–variance and 1/N strategy. Contribution/Results: Our approach explicitly models and exploits negative-edge information to reduce systemic risk. Empirical evaluation on two real-world financial datasets demonstrates that portfolios constructed from the selected subsets achieve superior or comparable risk-adjusted returns (e.g., Sharpe ratio), lower volatility, and enhanced robustness relative to full-universe benchmarks—validating the effectiveness and practical utility of signed networks for modeling financial risk structure.

Technology Category

• Reasoning under Uncertainty: Stochastic Optimization
• Search and Optimization: Non-convex Optimization
• Constraint Satisfaction and Optimization: Constraint Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Social Networks and Social Media: Social media analysis through the lenses of networks

Abstract

In this work, we consider weighted signed network representations of financial markets derived from raw or denoised correlation matrices, and examine how negative edges can be exploited to reduce portfolio risk. We then propose a discrete optimization scheme that reduces the asset selection problem to a desired size by building a time series of signed networks based on asset returns. To benchmark our approach, we consider two standard allocation strategies: Markowitz's mean-variance optimization and the 1/N equally weighted portfolio. Both methods are applied on the reduced universe as well as on the full universe, using two datasets: (i) the Market Champions dataset, consisting of 21 major S&P500 companies over the 2020-2024 period, and (ii) a dataset of 199 assets comprising all S&P500 constituents with stock prices available and aligned with Google's data. Empirical results show that portfolios constructed via our signed network selection perform as good as those from classical Markowitz model and the equal-weight benchmark in most occasions.

Problem

• Optimizing portfolios using weighted signed financial networks
• Reducing portfolio risk through negative correlation edges
• Selecting assets via discrete optimization on signed networks

Innovation

• Signed network models reduce portfolio risk
• Discrete optimization scheme selects assets
• Time series of signed networks built